应用初诊鼻咽癌磁共振成像结构化报告数据建立鼻咽癌复发预测模型  

Establishment of recurrence prediction model of nasopharyngeal carcinoma using structured MRI report data of newly diagnosed nasopharyngeal carcinoma

作　　者：王晓怡 谢辉 周健[1] 崔春艳[1] 刘立志[1] WANG Xiao-yi;XIE Hui;ZHOU Jian;CUI Chun-yan;LIU Li-zhi(Department of Imaging,Sun Yat-sen University Cancer Center,Guangzhou 510060,Guangdong,China)

机构地区：[1]中山大学肿瘤防治中心影像科,广东广州510060

出　　处：《川北医学院学报》2021年第9期1139-1145,共7页Journal of North Sichuan Medical College

摘　　要：目的:利用初诊鼻咽癌磁共振(MRI)结构化报告的特征建立鼻咽癌复发预测模型,并验证其效能。方法:回顾性分析792例接受调强放疗的非转移性鼻咽癌患者的临床及影像学资料,建立结构化报告并收集患者的MRI特征,并按照2∶1分为训练队列(n=528)和验证队列(n=264)。通过Cox回归分析、最小绝对收缩及选择算子(LASSO)、逐步回归分析筛选鼻咽癌预后的风险因素;对临床数据及T/N分期进行单、多因素分析以筛选独立预后因素。预后研究终点为局部-区域无复发生存(LRFS)。利用筛选出的结构化报告特征和TNM分期分别建立列线图预测模型。由结构化报告系统特征建立的预测模型预测准确度和判别能力通过一致性曲线确定,并与基于TNM分期系统建立的预测模型进行预测准确度及判别能力的比较。训练队列所建立的预测模型通过验证队列研究得到验证。结果:基于治疗前MRI的结构化报告系统特征构建的鼻咽癌复发列线图具有较好的诊断效能,且训练队列与验证队列有较好的校准力。训练队列和验证队列的C-index分别为0.730(95%CI:0.666~0.794)和0.654(95%CI:0.539~0.768)。与基于TNM分期构建的预测模型相比,两组间的差异有统计学意义(P=0.02)。基于结构化报告系统构建的列线图有更好诊断效能和校准力。结论:基于初诊鼻咽癌MRI结构化报告特征构建的预测模型在预测鼻咽癌预后方面有较高的诊断效能。在临床诊断工作中采用结构化报告,可以提高影像报告的可读性,同时提高预测鼻咽癌患者是否复发的准确性。Objective:To establish a prognostic model of nasopharyngeal carcinoma(NPC)based on the pre-treatment MRI structured reporting system and to verify its effectiveness.Methods:792 consecutive non-metastatic NPC patients treated with intensity-modulated radiotherapy(IMRT)were enrolled,retrospectively.Clinical data and MRI features from structured reporting system of the cohort were collected.Patients were randomly assigned to training(n=528)and internal validation(n=264)cohorts at a 2∶1 ratio via computer software-generated random numbers.Cox proportional hazards regression method,least absolute shrinkage and selection operator(LASSO)method regression model and stepwise regression analysis were performed for building prognostic models based on structured reporting system.Univariable and multivariable Cox proportional hazards regression method were used for independent prognostic factors selection of clinical data and TNM staging system.The endpoints of our study was loco-regional relapse-free survival(LRFS).We built nomograms based on MRI features of structured reporting system.The predictive accuracy of our nomogram models were determined by concordance index(C-index),and were compared with the nomogram models based on tumor-node-metastasis(TNM)staging system.The prediction model established by the training cohort was verified by the validation cohort study.Results:The nomograms based on structured reporting system’s features exhibited a good prediction performance.The C-index for training cohort and test cohort were LRFS:0.730(95%CI:0.666-0.794)and 0.654(95%CI:0.539-0.768),respectively.Significant differences was observed between the predictive models based on the structure reporting system and TNM staging system(P<0.02).The nomogram based on structured reporting system had better diagnostic efficiency and calibration power.Conclusion:The prognostic model based on the pre-treatment MRI structured reporting system features showed a satisfactory prediction performance.The use of structured reporting system in cli

关 键 词：鼻咽癌 磁共振成像 结构化报告 TNM分期系统 预测预后 

分 类 号：R739.6[医药卫生—肿瘤]